Provider file
A normalized profile for identity, hardware claims, uptime, benchmark samples, pricing, and workload limits.
Agent-native compute orchestration
Pre-flight checks for agent compute.
Proof Grid sits before execution. It scores providers, prices the route, records the checks, and asks for approval before an agent spends compute.
- Check
- Provider, hardware, policy
- Quote
- Cost and fallback route
- Hold
- No run before approval
Control plane, not marketplace.
Proof Grid makes compute supply readable before any job is sent.
System
The part between intent and execution.
Agents can request compute quickly. The harder problem is deciding which provider is acceptable, what the run will cost, and whether the user has approved it.
Plan object
A pre-execution route with selected provider, cost estimate, fallback, checks performed, and approval state.
Receipt hash
A compact record that ties the plan, provider, approval state, and execution metadata into an auditable artifact.
Planning ledger
A job should have a paper trail before it runs.
jobinference.embed.batch
providernode-7c9
trust94 / 100
quote$0.38
fallbackedge-cpu-grid
stateapproval_required
Trust receipts
Receipts are the output, not the decoration.
Each planned run gets a record with provider context, approval state, quote, and hash. The goal is boring accountability.
Provider
node-7c9.base
Hardware
H100 class GPU
Uptime
99.3% / 30d
Quote
$0.42 / task
Result hash
0x8f2...19a
Provider network
Provider claims should be comparable.
The table uses sample data. In production, rows come from signed provider profiles, monitoring, and verifier checks.
Provider
Best for
Trust
Quote
Northstar H100 Pool
Low-latency inference
94
$0.38
Vector Forge Cluster
Fine-tuning jobs
91
$12.80
EdgeMesh CPU Grid
Batch workloads
88
$2.14
Surplus adapter
Market routing after the plan is approved.
Proof Grid is adding a Surplus adapter in the public core repo. Surplus can provide live inference supply and price discovery; Proof Grid handles the pre-flight checks before that route is used.
01
Plan
Score providers, estimate cost, choose fallback.
02
Approve
Hold execution until the user or policy approves.
03
Route
Send approved inference work to a Surplus market.
04
Receipt
Attach market, model, settlement, and result metadata.
integration status
adapter scaffold live in proofgrid-core
Demo plan
Switch the workload. Watch the route change.
Route a short inference task to the fastest trusted GPU pool under budget.
Selected provider
Northstar H100 Pool
Trust score
94 / 100
Estimated cost
$0.38
Approval state
Ready for user approval
Live planning surface
The product should feel like this.
A narrow console for compute decisions: search, score, inspect, plan, hold. No run is sent until the policy says yes.
Proof Grid AI
A small guided assistant, not a chatbot.
The prompts below explain core behavior without pretending the site is connected to a live model.
Architecture
Four objects, one decision path.
Agent
Requests compute
Proof Grid
Plans and verifies
Provider
Runs approved job
Receipt
Returns proof
Roadmap
Ship the control plane first.
The first milestone is not a giant network. It is a reliable plan object that agents and users can inspect before compute is spent.
Agent-readable beta
Publish provider discovery, planning endpoints, metadata files, and example agent workflows.
Provider trust profiles
Add provider pages with hardware claims, uptime, benchmarks, policy limits, and pricing history.
Execution receipts
Return job status, logs, timing, result hashes, and provider context after each approved compute task.
Verification network
Expand toward third-party checks, reputation scoring, and automated fallback routing for agent workloads.
Protocol utility
$PROOF is for coordination work.
$PROOF is planned for places where the network needs accountability: provider bonds, verifier work, settlement paths, and reputation weighting.
Provider staking
Compute providers can bond $PROOF to publish trust profiles, accept jobs, and signal accountability for claimed capacity.
Verifier rewards
Independent checkers can earn $PROOF for validating provider claims, execution receipts, benchmark samples, and uptime signals.
Job settlement
Agents can use $PROOF for planning fees, receipt generation, and optional settlement paths when compute jobs are approved.
Reputation weighting
Staked participation and verified receipts can contribute to provider reputation, fallback routing, and dispute prioritization.
Utility design may evolve as the protocol is built. $PROOF is not required to understand the beta surface: discovery, planning, and receipts remain the core product.
Revenue model
Revenue comes from useful checks.
If the system helps teams avoid bad routes, bad providers, and missing receipts, it has clear paid surfaces.
Planning API fees
Developers and agent platforms can pay for provider search, trust scoring, route planning, and cost estimation at scale.
Provider verification
Compute providers can pay for verified profiles, benchmark checks, uptime monitoring, and richer trust metadata.
Receipt infrastructure
Teams can pay to store, query, and export execution receipts for compliance, debugging, billing, and agent audit trails.
Managed integrations
Larger customers can pay for private provider routing, custom policy rules, wallet approvals, and dedicated support.
Optional settlement fees can apply when Proof Grid coordinates approved compute jobs end to end. The product remains useful even before any settlement layer is enabled.
API
Small API, strict job boundary.
The public core repo starts with providers, plans, jobs, and receipts. Everything else can be added after the boundary works.
GET /providers
GET /providers/:id
POST /plan
POST /jobs
GET /jobs/:id{
"task": "inference",
"budgetUsd": 1,
"requirements": {
"gpu": true,
"maxLatencyMs": 1200
}
}Proof Grid
Make compute decisions inspectable.
The website is the surface. The core repo is the first implementation. Both point at the same rule: plan before execution.